Source-Term Prediction During Loss of Coolant Accident in NPP Using Artificial Neural Networks

Abstract

Nuclear power plant (NPP) is a highly complex engineering system which experiences a number of transients such as equipment failure, malfunctioning of process and safety systems, etc. during its operations. Such transients may eventually result in an abnormal state of the plant, which may have severe consequences if not mitigated. In case of such an undesired plant condition, the chances of the release of source term (e.g. release of Iodine, Caesium, Krypton, Xenon, etc.) and subsequent dose to public and to the environment cannot be neglected. The early knowledge of the expected release of source term will help in planning the emergency preparedness program. In view of this, several computational intelligence techniques have been studied and employed to early prediction of source term based on containment thermal-hydraulic parameters and also taking into account the actuation/non-actuation state of associated engineered safety features (ESFs). This paper presents an integrated framework based on artificial neural networks (ANNs) for early prediction of the expected release of source term during the large break loss of coolant accident (LOCA) in 220 MWe pressurised heavy water reactors (PHWRs). A simulated data of fission product release up to 48 h from the beginning of the LOCA has been considered in the model development. Several neural networks with forward and reverse configuration of the hidden layer were tried to reach at an optimal network for this problem. As the range of the input data was significantly large, a data transformation to a logarithmic scale was also performed to improve the efficiency and accuracy in prediction. The developed ANN model has been validated with the blind case LOCA scenarios. The performance of the final model is found to be satisfactory with a percentage error between actual and predicted value being under 5% expect for a few cases for all the species.